How I work
The same habits run through every project: decisions in writing, work in small scoped pieces, nothing ships without a check, and every mistake turns into a better process.
Plans with numbered, locked decisions
Every product starts from a written plan with numbered decisions and a reason for each. Changing a decision means updating the register and saying why, so the team never argues about what was agreed.
Scoped work orders with acceptance criteria
Work ships in scoped phases. Each work order lists requirements, what is in and out of scope, and acceptance criteria, and closes with a verification report. Done means every must-have requirement has a passing automated test and the full suite stays green.
Approval gates and green builds
Every deploy, customer-facing message, publication and payment change needs sign-off before it ships. Pull requests must pass CI before they merge, and automated test suites (PHPUnit and Pest, with static analysis) run on every product.
Blameless post-mortems
After any serious incident I write down the root cause and change the process, not just the code. A few examples:
| # | What happened | Process change |
|---|---|---|
| 1 | A server file-permissions change broke a live login page while internal checks still passed | Every permissions change now ends with a real web request check |
| 2 | A test run reached production data | A code guard blocks tests from ever running against production, plus a standing rule |
| 3 | A multi-tenant scoping bug was found in review | Fixed with a test that reproduces it, and CI must be green before any merge |
| 4 | Installing a second PHP version on a shared server broke a live app | A written install procedure for every server change |
Technical debt, decided in the open
I weigh technical debt against new features with the numbers on the table. On RepComet the plan's first locked decision was to extend the existing codebase rather than restart, because a working review engine with 271 tests already existed. On CrexiPay the opposite call was right: the first plain PHP version could not support payments safely, so it was rebuilt on Laravel with tests before new features were added.
Multi-server, queue-driven applications
I manage Laravel applications across separate production and staging servers, with Redis queues, Laravel Horizon, Supervisor-managed workers and scheduled jobs, plus event-driven Stripe and Twilio webhooks with retries and idempotency, so a retry can never double-charge a customer.
Clear communication for every audience
Engineers get specifics: requirements, risks and reproducible bug reports. Non-technical stakeholders get plain-language updates focused on what changes for them. I keep a daily briefing, an open-items tracker and weekly status reporting.
AI-assisted delivery, with people in charge
Much of the engineering on these products was AI-assisted under my direction, coordinating about 20 specialized AI agents across development, support, monitoring and SEO. I own the scope, requirements, decisions, acceptance and release, and anything public, financial or irreversible needs human approval.